AI safety researcher Daniel Kokotajlo says recent interpretability progress lowers the relative probability of catastrophic loss of control
He agreed tracking interpretability speed relative to model scaling is key
Story Brief
He agreed tracking interpretability speed relative to model scaling is key
